Technology and Tariffs" Operation: Coordinated influence campaign aimed at shaping US technology policy

OpenAI Deletes Accounts Tied to Chinese Propaganda on US Tech Policy

OpenAI has removed a cluster of accounts it believes originated in China that used its AI tools to generate online commentary and cartoons about US technology policy, tariffs, and trade restrictions. The move, disclosed through the company's regular reporting on misuse of its platform, adds to a growing list of covert influence operations that have been caught relying on commercial AI systems to produce content at scale.

What OpenAI found

According to OpenAI, the accounts in question were most likely operated by people based in the People's Republic of China. The operators used ChatGPT and related tools in two distinct ways: first, to draft comments that criticized US tech policies, including tariffs and export controls; and second, to generate cartoons and comic-style images mocking those same policies. The content appears designed to look like organic commentary from everyday internet users rather than coordinated messaging.

The accounts violated OpenAI's usage policies, which prohibit using its models to run influence operations or deceive people about the origin of content. OpenAI did not attribute the activity to any specific government body, describing the actors' origins as likely but not officially confirmed on behalf of any state institution.

Why AI makes influence operations easier

Coordinated campaigns to shape public opinion are not new, but generative AI has lowered the cost of running them dramatically. In the past, operators needed teams of fluent writers to produce persuasive text in their target language, and illustrators to create shareable visuals. Now, a small group can use large language models and image generators to produce hundreds of posts, comments, and cartoons quickly and cheaply.

OpenAI's report fits a broader pattern. The company and other AI providers have repeatedly disclosed and disrupted campaigns in recent years, some linked to Russia, Iran, and China, that used AI-generated text, images, and even videos to push political narratives. The stated US policies around semiconductors, AI chips, and trade tariffs are frequent subjects of such activity, given how consequential they are for China's tech sector.

How OpenAI detects and acts on this

OpenAI says it uses a combination of internal detection systems, human review, and tips from outside researchers to identify policy-violating activity. Once a suspicious cluster of accounts is identified, the company analyzes how the accounts used its models, looks for patterns linking them together, and then bans the accounts involved.

In this case, the company shared its findings publicly and, per its usual practice, coordinated with relevant platforms and researchers so the broader ecosystem could respond. Critics note that banning accounts only disrupts operators temporarily, since they can return with new credentials. Still, transparency about these campaigns helps researchers and platforms recognize and remove the resulting content elsewhere.

The wider context

The disclosure comes at a time of heightened tension between the United States and China over technology. US restrictions on advanced chips and AI-related exports, along with tariff measures, have become central points of friction. Online narratives around these policies, on both sides, have become a battleground for influence efforts.

OpenAI has positioned its threat reporting as part of an effort to prevent its tools from being weaponized in such campaigns. The company publishes these disruption notices periodically, documenting how its models were misused and what steps it took. Other major AI companies, including Anthropic and Google, have adopted similar practices, reflecting industry-wide concern that generative tools could amplify propaganda and disinformation.

At the same time, the episode highlights the limits of detection. AI-generated commentary, especially short social media posts, can be difficult to distinguish from genuine opinion. Cartoons and memes are similarly hard to police. OpenAI's action shows that platform-level detection is possible, but the resulting content may already have spread on social networks before accounts are cut off.

Who should use it / who should skip it

This report is most relevant for researchers, journalists, and policy watchers who track disinformation and foreign influence operations, as it offers a concrete example of how AI tools are being used in state-adjacent narrative campaigns. Platforms and trust-and-safety teams will also find the detection patterns useful. For ordinary ChatGPT users and businesses, there is little direct impact: the banned accounts were removed, and OpenAI has said its products remain available for legitimate use. Anyone hoping for a full technical breakdown of the campaign may be disappointed, as OpenAI's disclosures are necessarily limited to protect investigative methods.